Gesture Recognition For Vision-based Robot Demonstration Programming System

Human-Robot Interaction is gaining more and more attention in the field of robotics and industrial manufacturing.It is not only a hot issue at home and abroad,but also an inevitable trend in the development of robot technology in the future.And with the continuous development of robot technology,the application field of industrial robots has also been greatly developed and improved,and has begun to promote the application from the traditional manufacturing to the discrete manufacturing industry.Traditional manufacturing includes automobile manufacturing,machining,and discrete manufacturing industries include hardware,3C,and furniture industries.There are a large number of small parts assembly needs in the emerging manufacturing field.Because it is characterized by multiple varieties,small batches and short cycles,this field has extremely high requirements for the ease of use of industrial robots.Robot programming is the description of the sequence of actions that can be set by robots to complete specific tasks,and it is also the basis of improving the usability of robots.There are two common programming methods: offline programming and teaching programming.Teaching programming,also known as demonstration programming,refers to the sequence of the robot's actions for related tasks through manual guidance or operation of related devices.Because the teaching method is practical and easy to operate,most robots adopt it.The reason why robot demonstration programming has become a research hotspot in the field of robotics is that it can effectively simplify robot programming and can be used as an important way to improve the ease of use of robots.And gesture recognition is a more important part of the robot demonstration programming system.This article studies the gesture recognition in industrial robot demonstration programming.Considering the friendliness of the actual user(worker)operating system in a specific industrial environment,this article considers and designs a user-friendly recognition system from the user's perspective.It mainly includes the following:1.Customization of 6 gestures,which respectively represent 6 operations in the system,and use of Kinect for real-time point cloud information collection.In the segmentation gesture,depth threshold segmentation is used to remove unnecessary point cloud information.The use of depth information for collection can ignore the effects of light or background.For the needs of small and medium-sized factories,the applicability and ease of use of this method are better and more convenient.2.Projection of the point cloud of the gesture for the contour extraction.Use clustering and other methods to find the fingertips and other positions.Then use the defects in the gesture as features and use convex hull detection to find the defects.Finally calculate the number of fingers.The segmentation and recognition results of instruction gestures generated in the system and the successful robot demonstration programming system show that the algorithm is effective and practical.